ResearchBenchmarksImage Classification on MNIST ε=0.3 (test)Follow98.85Standard AccuracyAD-CERT98.444498.549798.65598.7603Jun 30, 2026Evaluation ResultsMethodMethodLinksStandard AccuracyCertified AccuracyAD-CERTε=0.3ε=0.32026.0698.8593.98MTL-IBPε=0.3ε=0.32026.0698.7493.9SABRε=0.3ε=0.32026.0698.6693.68TAPS/STAPSε=0.3, Selection proto...ε=0.3, Selection protocol=Stronger of TAPS and STAPS based on certified accuracy2026.0698.5693.95IBPε=0.3ε=0.32026.0698.5493.8SORTNETε=0.3, Architecture=No...ε=0.3, Architecture=Non-ReLU-based2026.0698.4693.4